Preface¶

DSH’s philosophy is “Everything is a plugin.” For developers who need to break long workflow tasks into observable, retryable, and auditable steps, pluggability is easier to control than one-off calls. dsh-video-studio (Whale/鲸影) integrates AI video and manhua production into DSH, providing a seven-stage directorial pipeline, multi-account quota pools, prompt self-optimization and self-auditing, with the goal of prioritizing quality over cost.

What is this¶

One-sentence positioning: DSH native AI video and manhua production plugin.

The owner in the repository path is hackerFish; the package.json package name is @hackerfish/dsh-video-studio. The version is 0.2.0, and the license is MIT.

It solves the problem of managing the long chain of story, script, storyboard, assets, video generation, and final editing within a single plugin, while retaining step-by-step execution, quota scheduling, prompt optimization, and audit records.

Seven-Stage Directorial Pipeline¶

Whale’s pipeline consists of seven stages:

story → script → storyboard → master asset → shot assets → video → final cut

The first three stages are LLM stages. Documentation states: the first three stages are executed by the DSH session model; the plugin does not call external LLMs. This means the current session model handles story, script, and storyboard, while subsequent assets, video, and editing are scheduled by the plugin.

Multi-Account Quota Pool and Credential Library¶

Whale provides multi-account quota pools/schedulers with capabilities including multi-account rotation, daily caps, exponential backoff, failure fallback, and complete audit logs.

The credential library location is:

~/.whale/whale.json

Documentation notes that the credential library uses 0600 permissions, atomic writes, and follows $DSH_HOME. API responses are masked. Account management UI is located in the “Whale Accounts” tab in plugin settings.

The account management interfaces are as follows:

GET /dsh-video-studio/accounts
POST /dsh-video-studio/accounts
DELETE /dsh-video-studio/accounts

These are used to list accounts (and mask them), add accounts, and delete accounts, respectively.

Prompt Engineering and Scoring Backwrite¶

Whale’s prompt engineering is divided into four layers: parameterized professional template library, composable quality enhancers, local prompt optimization, and scoring backwrite.

Common tools:

whale_optimize_prompt
whale_quality_review

whale_optimize_prompt is used to upgrade draft prompts to pro-grade prompts locally. whale_quality_review is used for rule-level QC. Scoring backwrite combines shot scores with enhancers and writes them back to records for the optimizer to select based on historical performance.

Style Memory and Consistency Assets¶

Whale has a style genome/memory: style DNA, shot template score evolution, retry feedback, preserved across sessions.

The consistency asset library contains character/scene master assets and per-shot variants, automatically injecting reference images.

Parallel Shots¶

Parallel shots support batch submission, concurrent polling, and configurable concurrency.

Provider Adaptation and Status¶

The providers listed below come from verified sources. Real-time availability may depend on real keys, GPU, or platform status, not all verified.

Final Cut and Dubbing¶

Whale’s final cut capabilities include ffmpeg auto-rendering, JianYing draft export, optional lip-sync, and say TTS or external voiceFile.

Lip-sync is an optional capability; it falls back to the original clip on failure without blocking rendering. Shots can also carry pre-recorded voiceFile to skip local TTS.

The distribution package targets 4 Chinese platforms, providing spec and compliance pre-checks.

DSH Integration Interface¶

Model tools:

whale_story_presets
whale_storyboard
whale_generate_video
whale_optimize_prompt
whale_quality_review
whale_comfyui_workflow
whale_self_audit

Host routes:

/dsh-video-studio/health
/dsh-video-studio/runs
/dsh-video-studio/accounts

Client UI:

鲸影 (Status)
鲸影工作台
鲸影账号

Typical Usage¶

Listed below by reproducibility.

1. Preset Manhua Package Mock Run¶

Documentation notes support for 5 genres, supports whale_story_presets and scripts/demo-presets.ts end-to-end mock running.

scripts/demo-presets.ts uses a mock provider with zero keys to run the preset package.

2. Convert preset to pipeline-ready script¶

Use:

whale_story_presets

To convert a preset id to a pipeline-ready script.

3. Local prompt optimization¶

Use:

whale_optimize_prompt

To upgrade draft prompts to pro-grade prompts.

4. Execute provider routing/video generation¶

Use:

whale_generate_video

5. Rule-level QC¶

Use:

whale_quality_review

6. Generate ComfyUI workflow JSON¶

Use:

whale_comfyui_workflow

7. Self-audit¶

Use tool:

whale_self_audit

Or run:

npm run self-audit

It scans the repo and writes to:

docs/AUDIT-REPORT.md

8. Account management¶

Add, list, and delete accounts via UI or host routes:

POST /dsh-video-studio/accounts
GET /dsh-video-studio/accounts
DELETE /dsh-video-studio/accounts

Installation and Enablement¶

Verified sources do not provide official install commands. To avoid name concatenation, commands like dsh plugin add ... are not written here.

Verifiable package info:

{
  "name": "@hackerfish/dsh-video-studio",
  "version": "0.2.0",
  "license": "MIT"
}

Before installing, confirm the plugin directory page or installation instructions provided by the repository. The plugin runs with the current dsh process permissions; check the source code and license before installing.

Applicable Scenarios and Notes¶

Suitable for scenarios requiring the integration of AI video/manhua generation workflows into DSH, especially where step-by-step execution, multi-account quota scheduling, prompt optimization, audit logs, and local editing exports are required.

Notes:

• Some providers have not been tested with real keys or real GPUs; real-time availability depends on platform status.
• DashScope wan video is billed at ¥0.70/s (1080p), not a free quota.
• tongyi-wanx free tier is text-to-image; video needs membership.
• jimeng’s text-to-video queue may remain SystemBusy; the current free route in documentation is wanx images → image-to-video.
• Lip-sync is an optional capability; it fails back to the original clip and does not block rendering.

Links¶

• Plugin directory page: https://www.skillhub.cn/plugins/hackerFish/dsh-video-studio
• GitHub: https://github.com/hackerFish/dsh-video-studio